Abstract
Crime forecasting is beneficial in providing useful information to authorities in planning effective crime prevention measures. The two types of analysis used in crime forecasting are univariate and multivariate. Comparatively, multivariate analysis provides better forecasting accuracy because of its ability to discover crime patterns not previously seen. Crime is strongly influenced by several external factors, including economic, social and demographic. Hence, an analysis is needed to identify and select relevant factors that influence crime and can later be used to improve forecasting accuracy. Neighborhood Component Analysis (NCA) is a reliable form of analysis for identifying significant relationships between factors and crime data. Several model types have been introduced in crime forecasting, including statistical and artificial intelligence models. Recently, the artificial intelligence model has come into favour because of its ability to handle nonlinearity patterns in crime data well. Within the artificial intelligence model, Gradient Tree Boosting (GTB) shows good performance as it produces a robust and reliable forecast result. GTB uses least square function as a loss function for error fitting during training. Findings show that, in addition to using least square function, implementing other standard mathematical functions that fit to the crime data increases forecasting accuracy. In other cases, both NCA and GTB are sensitive to parameters input. Dragonfly Algorithm (DA) is a promising, nature inspired metaheuristic algorithm that is capable of solving such problems.
Highlights
Forecasting is a means of predicting or making a statement regarding uncertain events, based on past or present knowledge under specific criteria
The results show that the Autoregressive Integrated Moving Average (ARIMA) model outperforms the standard statistical model when it fits the data well and further enhances forecasting accuracy
Univariate analysis involves only single time series data to develop the forecasting model, while multivariate analysis employs more than one set of time series data in model development
Summary
Forecasting is a means of predicting or making a statement regarding uncertain (future) events, based on past or present knowledge under specific criteria. Most artificial intelligence models are sensitive to parameters Such problems have been addressed by several researchers, especially in ANN (Yeh, 2013; Rather et al, 2017; Hipp and Yates, 2011) and SVR (Wu et al, 2007; Wu et al, 2009; De Oliveira and Ludermir, 2014). These researchers have applied a promising solution to such problems by integrating other artificial intelligence techniques into existing artificial intelligence models, developing a new hybrid model (Wu et al, 2009; De Oliveira and Ludermir, 2014; Hipp and Yates, 2011) Such approaches have proven to be successful because the hybrid model, which is more robust to changes in time series patterns, is able to outperform other singular models (Cook and Durrance, 2013)
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
